This is a great visualization of free data, although not the first in this style, but it's usefulness in actual forecasting or nowcasting is rather limited.
Interpolation between sparse grid points can result in missing fine details, like the subtle boundaries that kick off the most violent storms in the central Plains.
Limiting to just GFS and GEM make sense from a proof of concept level, however these are long range models that play in the 10-16 day range. GFS in particular uses a 13km spaced grid that isn't convection allowing, meaning it can't model individual storms well. GFS is typically only output every 6 hours as well so it can easily get out of sync on forecasts for the day of.
It would be great to see these types of visualizations incorporate something fast and higher resolution like the HRRR or even one of the NAM/WRF 4km variants, but that is a lot more data than what is currently being ingested.
The best weather information (for US citizens) hands down is still your local NWS office. I'd recommend everyone bookmarking their site and following them on social media.
just returned from the canaries where windytv proficiently helped me to pick my bikeroutes, adapted to the prevailing wind conditions.
these winds change fast and seem unpredictable and although the connection between general weather and wind is somehow limited it seems clear to me that it must be hard to make any accurate predictions.
I used to be married to a military forecaster. It was always interesting to hear about how this stuff works, and it seems to me that "models not fitting reality" comprised at least 50% of their office drama. Forecasters have their own preference for models and "past experience" which leads them to very different conclusions. And climate change is making these models much less effective over time, adding even more excitement.
In my region (Portugal) their predictions regarding rain on 2-3 days are correct enough to make people come and ask me.. and the temperatures are optimized towards 'mildy'. E.g: You see 30C, count with 32-33C; 5C expect 3C. Note that only their GFS 27km model is updated and trustable on free mode.
> This is a great visualization of free data, although not the first in this style, but it's usefulness in actual forecasting or nowcasting is rather limited.
Indeed, it reminds me of this, which has been around for years:
Will GOES-16 improve the existing models? Or is the plan to create new models? I'm really curious to know how the higher resolution images will be used.
Current models ingest new data from a variety of sources, including surface observations, buoys, airplanes, and GOES-derived data. The derived data might be more accurate and might be used more extensively going forward - I'm not entirely sure.
IMO (as an amateur) the bigger impact is for convective meteorologists that are continually watching satellite imagery and the work that places like CIMSS are doing in analyzing satellite imagery and detecting patterns indicative of severe weather. These detection algorithms will have a higher degree of confidence and can be triggered several minutes earlier now - possibly providing earlier warning for tornadoes.
We have some internal models that will benefit from the new GOES 16 data. Particularly the cloud mask product and the higher spatial/temporal resolution will be interesting. We develop our own internal cloudmasks using a custom tool and I'm interested to see how they differ. In anticipation of the higher spatial/temporal resolution data, we're updating our tools to reduce the memory footprint.
GOES-16 will provide the data to allow a significant improvement in forecasts, especially of extreme weather events. Because it can have a variable scan pattern, it can do wide area scans and higher frequency scans tracking storms. The new lightning sensor allows better measurement of storm intensity. It also has finer discrimination for spectral information, 4x increase in resolution, etc.
Interpolation between sparse grid points can result in missing fine details, like the subtle boundaries that kick off the most violent storms in the central Plains.
Limiting to just GFS and GEM make sense from a proof of concept level, however these are long range models that play in the 10-16 day range. GFS in particular uses a 13km spaced grid that isn't convection allowing, meaning it can't model individual storms well. GFS is typically only output every 6 hours as well so it can easily get out of sync on forecasts for the day of.
It would be great to see these types of visualizations incorporate something fast and higher resolution like the HRRR or even one of the NAM/WRF 4km variants, but that is a lot more data than what is currently being ingested.
The best weather information (for US citizens) hands down is still your local NWS office. I'd recommend everyone bookmarking their site and following them on social media.